Aston Villa vs Arsenal: when the draw is more likely than the home win
By FootInsights · Published · Updated 25 Sep 2026 · 5 min read
Most of the time a football model’s three numbers arrive in a familiar order: favourite, second team, draw. The draw is the outcome that almost never leads, and it is rarely even second. So it is worth stopping when the draw is more likely than the home win — which is exactly what our model says about Aston Villa vs Arsenal at Villa Park on Monday night. In our 26 August model run the fixture prices at 26.3% Aston Villa, 26.7% draw, 47.0% Arsenal. Villa are not the favourite and they are not the second most likely outcome either. They are third, at home.
What it takes for a home side to be third at its own ground
The gap is 0.4 of a percentage point, which is not a meaningful edge on its own — nobody should read it as “the draw is the bet”. What matters is the ordering, because of how the ordering is produced.
In a goals model, each club’s attack and defence are estimated from the history of scorelines, a home-field term is applied, and the resulting goal expectations become a distribution over scores. The draw probability in that framework is not a free parameter. It falls out of the score grid, and it peaks when two sides have similar, low goal expectations — the 0-0, 1-1, 2-2 diagonal gets fat when neither team is expected to score much.
That is not this game. Our model has the same run at 51.2% for over 2.5 goals and 54.8% for both teams to score, which is an ordinary, mildly goal-friendly fixture rather than a grinding one. The draw here is not high because the game looks tight. The draw is second because the home win has been pushed below it, and pushing a Premier League home side below the draw takes a genuinely poor reading of the home side relative to the visitor.
Where Villa’s rating actually comes from
Two things, both of which the model reads as goals.
The first is what happened on opening weekend. Villa lost 0-4 at Brighton, as Sky Sports’ fixture page records, while Arsenal beat Coventry City 3-0 and had already taken the Community Shield 3-0 against Manchester City. To a margin-sensitive model, a four-goal defeat is not one result — it is a large downward revision of how well a side defends, and it was the single heaviest result Villa had contributed this season.
The second is the longer record behind it. One weekend cannot move a rating that far on its own. Villa arrived at 26.3% because a season’s worth of scorelines had already left them well short of Arsenal, and matchday 1 widened the gap.
Note what this did not include. The model had no column for the fact that, per Sports Mole’s team-news report, Youri Tielemans, Morgan Rogers and Ezri Konsa had all left the club that summer. The model’s view of Villa was earned by results, and the results were earned by a squad that had since been partly dismantled. If you think those departures matter, you should read the 26.3% as an upper bound on Villa rather than a fair estimate — and that judgement was the reader’s to make, not the model’s, because it could not see a transfer.
The suspension the model cannot see, and what it is worth
The same Sports Mole report confirms two absences for Monday. João Gomes is suspended after “a straight red card for kicking out against Georginio Rutter” at Brighton, and Amadou Onana is “set for a considerable period on the treatment table due to a serious knee injury”.
Neither of those is in the 26.3%. That is a limitation, and it is one we would rather state than paper over: the model at the time read results, not team sheets, and its probability should be read as “what the scoring record implied”, with team news applied by you on top. Our how our predictions work page describes the model, and the public track record is scored on exactly the numbers the model published, absences included in neither direction.
What is interesting is the direction of the correction. Losing a suspended midfielder and an injured one does not simply shave points off the home win and hand them to Arsenal. In a goals model, weakening a side’s midfield lowers its own scoring rate and raises its opponent’s — which lowers the home win, raises the away win, and pulls the draw down too, because the score grid’s diagonal thins when the two goal expectations separate. A reader applying that correction by hand should expect the draw to lose its narrow lead over the home win, not extend it. The ordering we published is the ordering before team news, and it is the most fragile part of the forecast.
How to read a 26.3 / 26.7 split
Honestly: as “these two outcomes are indistinguishable, and the third one is much likelier than either”. A 0.4-point separation is inside the noise of any model fit, and anyone quoting it as a finding is over-reading it. The finding is the 47.0% away figure and the fact that a home side at Villa Park could not get clear of a draw.
For what it is worth, we are not alone in that ordering. An editorial forecast published on 12 August put the game at Arsenal 43%, draw 30%, Villa 27% — a wider draw-over-home gap than ours, from an independent starting point. That is corroboration of the shape, not of the exact numbers, and it was written before the red card at Brighton.
The live probabilities, the goals lines and whatever prices our feed carries by kick-off are on the match page, which re-renders at every build; the figures quoted in this article are frozen at the 26 August run. If you want the wider picture of how the model reads this division right now, the Premier League page has every matchday 2 fixture, and Aston Villa’s team page carries the run of results that produced the numbers this article has been arguing with.